Scholarlink

<p>The physicochemical properties of chemicals are essential for understanding their behaviour within biological systems and their environmental impacts. With the rapid increase in the synthesis and release of chemicals into the environment, there is an urgent need for efficient screening methods to characterize these substances. This understanding is fundamental for developing safety guidelines and regulatory policies. In this study, we developed machine learning (ML)-based quantitative structure-property relationship (QSPR) models to predict key physicochemical properties, including vapor pressure (LogVP), octanol-water partition coefficient (LogP), melting point (MP), boiling point (BP), and water solubility (LogS) using a collected dataset of 14,207 diverse organic chemicals. We employed a stacked ensemble method to generate final predictions, using four commonly used ML algorithms as base learners: random forests, extreme gradient boosting, support vector machines, and multi-layer perceptrons. We used partial least squares as a meta-learner. The developed stacked models demonstrated strong predictive performance for both the training and test sets. The determination coefficients (R<sup>2</sup>) for all models ranged from 0.936 to 0.986, while the predictive R² values (Q<sup>2</sup><sub>F1</sub>) fell between 0.807 and 0.959. Finally, we made our stacked models accessible to others by developing a Python-based expert system that provides predictions along with their reliability.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Development of stacked ensemble models for estimating the physicochemical properties of organic chemicals

  • 期刊论文

  • 发表日期：

    2026年2月26日

查看全文
  • Souvik Pore,
  • Kunal Roy

摘要

The physicochemical properties of chemicals are essential for understanding their behaviour within biological systems and their environmental impacts. With the rapid increase in the synthesis and release of chemicals into the environment, there is an urgent need for efficient screening methods to characterize these substances. This understanding is fundamental for developing safety guidelines and regulatory policies. In this study, we developed machine learning (ML)-based quantitative structure-property relationship (QSPR) models to predict key physicochemical properties, including vapor pressure (LogVP), octanol-water partition coefficient (LogP), melting point (MP), boiling point (BP), and water solubility (LogS) using a collected dataset of 14,207 diverse organic chemicals. We employed a stacked ensemble method to generate final predictions, using four commonly used ML algorithms as base learners: random forests, extreme gradient boosting, support vector machines, and multi-layer perceptrons. We used partial least squares as a meta-learner. The developed stacked models demonstrated strong predictive performance for both the training and test sets. The determination coefficients (R2) for all models ranged from 0.936 to 0.986, while the predictive R² values (Q2F1) fell between 0.807 and 0.959. Finally, we made our stacked models accessible to others by developing a Python-based expert system that provides predictions along with their reliability.